Machine Performance Monitoring and Fault Classification using Vibration Frequency Analysis

Machine anomalies in manufacturing directly affect the production yield and factory operation efficiency if such anomalies cannot be detected in time. Real-time monitoring of machine health condition not only improves machine throughput by reducing unplanned downtime caused by machine failure but also saves cost for unnecessary routine maintenance. This paper presents a systematic approach for real-time or near real-time machine performance monitoring solution development from data collection, feature extraction, data analytics to real-time machine fault and machine status classification. Three data-driven machine-learning approaches using one vibration sensor data are proposed to detect two common machine failure modes during machine turning process. To evaluate the the performance of each approach, three machine-learning algorithms (Random Forest, K Nearest Neighborhood, and Support Vector Machine) are implemented and tested. Evaluation results on the actual machine data shows that a two-layered classification structure with random forest algorithm as the base has high classification accuracy on the machine status including machine fault detection. The developed data-driven machine health monitoring solution is deployed in the IoT device for real-time data collection and processing and results are sent data server for data visualization.